Ultrasound Feature Evaluation for Robustness to Sensor Shift in Ultrasound Sensor Based Hand Motion Recognition

Abstract

Pattern Recognition based approaches have offered great promise in the field of bio-signal controlled prosthesis. Traditionally Surface Electromyography based Approaches (SEMG) have been used to satisfy the purpose of providing Bio-Signal control in upper extremity Prosthesis. Although these methods have been shown to be robust, there still exists issues in performance within clinical environments. In recent years, Ultrasound signal based methods have seen growing interest within the field of motion Recognition, largely due to the increased resolution, deeper muscle observation, and reduced cross-talk that can be achieved in comparison to SEMG methods. However, the methods to be applied for hand Motion recognition are still only just beginning to be explored. In this paper, we shall investigate the applicability of SEMG feature extraction techniques to Ultrasound based hand motion recognition and the subsequent impact of Sensor shift on these features. The results of this study indicate that SEMG feature extraction techniques have excellent single location accuracy in Ultrasound based Hand motion recognition. However this paper more visibly presents the strong impact of Sensor Shift on A-Mode ultrasound based hand motion Recognition, and finally presents which feature extraction methods are most robust to this shift.